1 μm C-Doped GaN Thin Buffer on Sapphire with >3 kV Lateral Breakdown Voltage Grown by MOCVD

Abstract

Sapphire-based GaN buffers face inherent challenges from the lattice mismatch between GaN and sapphire, which leads to high threading dislocation density and limits lateral breakdown voltage. In this work, we investigated the optimization of metal–organic chemical vapor deposition (MOCVD) growth parameters—specifically carbon doping concentration, GaN buffer thickness, AlN nucleation layer thickness, growth pressure and V/III ratio—to enhance crystal quality and breakdown performance. A sapphire-based C-doped GaN buffer layer was successfully fabricated exhibiting a lateral breakdown voltage exceeding 3000 V across a 30 μm electrode gap corresponding to an average breakdown electric field of approximately 1.0 MV/cm, accompanied by low threading dislocation density, excellent surface roughness and low leakage currents. This study provides technical insights and practical growth guidelines for high-voltage sapphire-based GaN buffer layers, establishing the material basis for future high-voltage power device applications.

1. Introduction

Driven by both the global energy transition and advancements in semiconductor technology, power devices are rapidly evolving toward higher frequencies, greater efficiency, smaller form factors, and enhanced reliability [1]. They are widely used in core sectors such as new energy vehicles, data center power supplies, and smart grids, becoming critical components that underpin the energy-saving requirements and the upgrade of next-generation electronic equipment [2]. As the demands for power density and switching frequency in power electronic systems continue to rise, traditional silicon (Si)-based power devices are constrained by their inherent material properties. Key parameters such as breakdown electric field and electron mobility have approached theoretical limits, making it difficult to meet the application requirements of higher-frequency, higher-power scenarios [3]. Consequently, the development of new wide-bandgap semiconductor materials and corresponding devices has become an inevitable trend in the industry [4].

As a leading wide-bandgap semiconductor, gallium nitride (GaN) outperforms silicon in power electronics, enabling high-efficiency, high-density power conversion beyond the intrinsic limits of silicon-based devices. GaN high-electron-mobility transistors (HEMTs), which leverage the high-mobility two-dimensional electron gas (2DEG) at AlGaN/GaN heterojunctions, have achieved broad commercial adoption [5,6,7,8].

However, substrate selection is critical for high-voltage GaN devices: conductive Si substrates force the buffer to take the full voltage drop, causing parasitic coupling and necessitating very thick buffers to achieve high breakdown voltage (V_BR); SiC substrates, while excellent in thermal conductivity and vertical blocking, suffer from high substrate and processing costs that hinder large-scale, cost-sensitive adoption. Sapphire, as an intrinsically insulating substrate, offers inherent advantages in suppressing vertical leakage, reducing substrate coupling, and simplifying high-resistivity buffer and lateral field management, while benefiting from the LED-driven mature supply chain that provides large-size, low-cost wafers and established processing [9,10,11,12]. The main challenges are high dislocation density due to large lattice mismatch with GaN and relatively low thermal conductivity of sapphire, but these issues can be mitigated by epitaxial engineering (e.g., AlN nucleation optimization, graded buffers, and doping strategies), advanced packaging, and hybrid substrate/system-level thermal management [13,14,15,16]. Therefore, systematic development of GaN power devices on sapphire can provide a competitive trade-off among performance, cost, and manufacturing scalability, and holds significant practical value for mid-to-high voltage, low-to-moderate average power, and applications requiring strong electrical isolation.

Many teams have already achieved high-voltage GaN HEMTs on sapphire substrates. K.T. et al. combined junction isolation (JI) and dielectric isolation (DI) in a hybrid RESURF technique to fabricate E-mode GaN HEMT devices with a p-GaN bottom layer [17]. By optimizing the Mg doping concentration, they achieved a breakdown voltage of 1300 V for the devices. L.X. et al. successfully fabricated high-performance D-mode GaN HEMT devices by preparing a 1.5 μm buffer layer and three field plates, achieving a breakdown voltage of 1700 V on 6-inch sapphire substrates [11]. Utilizing a 100 nm undoped composite ultra-thin buffer layer and a field plate structure, L.S. et al. fabricated an E-mode GaN HEMT with a breakdown voltage reaching 2300 V by adjusting the distance of the L_gd [18].

Existing research often concentrates on the device level, seeking to increase the breakdown voltage by incorporating field plates and other complex structures, which inevitably increases process complexity and manufacturing costs. Actually, buffer leakage is a critical issue in GaN high-voltage devices, as residual electrons in the GaN buffer can form leakage paths under high electric fields. Carbon doping is commonly used to compensate these residual electrons through acceptor-like states, thereby increasing the buffer resistivity and improving the blocking capability of the device. However, most of the buffer optimizations focus on individual growth parameters, which may overlook the complex interactions among multiple MOCVD growth variables. A comprehensive approach is needed to concurrently manage critical parameters such as carbon doping concentration, GaN buffer layer thickness, AlN nucleation layer thickness, growth pressure and V/III ratio to synergistically improve both crystal quality and electrical performance.

In this work, we propose a multi-parameter co-optimization route within the investigated growth window for GaN buffer layers on sapphire substrates, in which carbon doping concentration, buffer layer thickness, AlN nucleation layer thickness, growth pressure, and V/III ratio are systematically tuned in a coupled manner. Rather than treating these parameters independently, we demonstrate that the breakdown performance is governed by the interplay among (i) dislocation density, (ii) carbon-induced compensation, and (iii) surface morphology, which together determine the dominant leakage mechanisms and electric field distribution. Through this approach, we identified a critical optimization window that balances defect suppression and compensation efficiency, and fabricated cGaN buffer structures with breakdown voltage over 3000 V, low dislocation density and excellent surface morphology. This work is promising for the C-doped GaN buffer on sapphire and test structure configuration used here, and provides a useful experimental reference for similar sapphire-based C-doped GaN buffer structures.

2. Materials and Methods

The GaN buffer layers were epitaxially grown on 4-inch c-plane (0001) sapphire substrates using an AIXTRON G2 planetary MOCVD system. Prior to growth, the sapphire substrates underwent in situ cleaning in ambient H₂ to remove surface contaminants. The AlN nucleation layer was deposited first in two sequential stages at middle temperatures (MT), followed by a 500 nm GaN buffer layer at a growth temperature of 1050 °C. The surface exhibits a smooth and uniform morphology without noticeable cracks or defects. Trimethylgallium (TMGa), trimethylaluminum (TMAl), and ammonia (NH₃) were employed as the Ga, Al, and N precursors, respectively. Nitrogen (N₂) served as the carrier gas throughout the growth process.

A comprehensive series of samples was designed and fabricated, comprising 12 main samples across three distinct parameter optimization groups, along with 2 supplementary samples specifically grown under reduced pressure and lowered NH₃ flow conditions. The detailed nominal carbon concentration, cGaN buffer thickness, AlN nucleation layer thickness and calculated V/III ratio are summarized in

Table 1. Epitaxial parameters for samples A1–D2. The V/III ratios were 1739 and 870 at lower growth pressures for samples D1 and D2, respectively.

Sample	Nominal C Concentration (cm−3)	AlN (nm)	cGaN (μm)
A1	8 × 1017	400	0.50
A2	1.61 × 1018	400	0.50
A3	4.69 × 1018	400	0.50
A4	9.38 × 1018	400	0.50
A5	1.34 × 1019	400	0.50
B1	4.69 × 1018	400	0.25
B2	4.69 × 1018	400	0.50
B3	4.69 × 1018	400	1.00
B4	4.69 × 1018	400	1.50
C1	4.69 × 1018	320	1.00
C2	4.69 × 1018	480	1.00
C3	4.69 × 1018	640	1.00
D1	4.69 × 1018	480	1.00
D2	4.69 × 1018	480	1.00

.

The first optimization group (samples A1–A5) investigated the effect of C doping concentration on buffer layer resistivity and leakage suppression. C doping is a well-established technique to achieve semi-insulating GaN buffers by compensating residual donors and forming deep acceptors, thereby reducing leakage currents and enhancing breakdown voltage [19,20,21,22,23]. To systematically elucidate this effect, five samples were designed with all other epitaxial parameters held constant while the nominal C doping concentration was varied across the range of 8 × 10¹⁷, 1.61 × 10¹⁸, 4.69 × 10¹⁸, 9.38 × 10¹⁸, and 1.34 × 10¹⁹ cm⁻³. In this study, C₂H₄ was used as a carbon dopant, and different concentrations of carbon doping were achieved by adjusting the flow rate of C₂H₄. It should be noted that the specific carbon concentrations discussed in this work are nominal values rather than direct post-growth measurements for the sample series. These nominal values were determined based on rigorous prior calibrations of our MOCVD reactor. Specifically, a C₂H₄ precursor flow rate (based on our previous studies) was selected for the growth of the cGaN layer under the same growth conditions, including growth pressure, temperature, and the flow rates of TMGa and NH₃. The resulting C incorporation concentration at this precursor flow was then measured by secondary ion mass spectrometry (SIMS). The systematic correlation between carbon incorporation and the MOCVD growth conditions was thus established using these reference samples. Different carbon doping concentrations in the present experiments were achieved by proportionally adjusting the dopant precursor flow according to this calibrated relationship. The second optimization group systematically probed the influence of the cGaN buffer thickness on structural quality and lateral breakdown behavior. We varied the cGaN thickness across four samples (explored range of 0.25~1.50 μm) while keeping the optimized carbon doping conditions constant. The motivation is that the buffer thickness controls the extent of dislocation bending and annihilation, the distribution of compensating carbon, and the distance over which leakage paths can develop toward the surface or device contacts. The third optimization group (samples C1–C3 and B3) focused on the effect of AlN nucleation layer thickness on buffer leakage and breakdown characteristics. The AlN nucleation layer plays a critical role in accommodating the lattice mismatch between sapphire and GaN and in initiating two-dimensional growth to suppress threading dislocation formation. Four samples with AlN thicknesses at the range of 320 to 640 nm were designed to explore this parameter space. The objective was to identify the optimal AlN thickness that minimizes threading dislocation density and surface roughness, thereby maximizing electrical blocking capability. The final optimization group (samples D1–D2) involved further refinement of cGaN growth conditions based on the optimal baseline (sample C2). Sample D1 was obtained by reducing the cGaN layer growth pressure from 300 mbar to 50 mbar, a modification intended to improve crystal quality by reducing precursor dissociation and point defect incorporation. Subsequently, sample D2 was derived from D1 by further decreasing the NH₃ flow rate (from 0.40 mol/min to 0.20 mol/min), effectively lowering the V/III ratio (from 1739 to 870). It was noted that the same nominal carbon concentration was targeted for C2, D1, and D2, but the actual incorporation efficiency may be influenced by the reduced pressure and V/III ratio. This group was employed as a supplementary data set to validate the trends observed in the main sample series, to explore potential synergistic effects between growth parameters, and to investigate how growth parameters regulate carbon incorporation efficiency, thereby influencing leakage current and breakdown characteristics.

Then a simplified lateral composite electrode was designed, with Ti/Al (10/500 nm) metal stacks directly fabricated on the cGaN buffer layer by magnetron sputtering (ULVAC ei-5z, Chigasaki, Japan) to quickly feedback for breakdown voltage testing, followed by metal lift-off to complete electrode preparation. The testing structure is shown in Figure 1. Then the sample’s OM (SOPTOP MX8R, Ningbo, China) image and electrode pattern is shown in Figure 2.

Structural characterization was performed using high-resolution X-ray diffraction (HRXRD, Malvern Panalytical X’Pert³ MRD, Almelo, The Netherlands) to assess crystal quality. Rocking-curve measurements of both symmetric (002) and asymmetric (102) reflections were conducted to determine screw and edge threading dislocation densities, respectively. Surface morphology was evaluated using atomic force microscopy (AFM, SEMILAB, Budapest, Hungary) in tapping mode over 5 µm × 5 µm scan areas to quantify root mean square (RMS) roughness values. Lateral electrical characterization was performed using a semiconductor parameter analyzer Keysight B1505A. Circular metal electrodes were patterned with different contact-to-contact gaps (L_gap) of 30, 50, 70, 90 and 110 μm, and we performed the measurements for the lateral breakdown voltage (V_BR) and leakage current using electrodes with a fixed distance of 30 µm (Type A), while Types B–E were included in the same mask layout as auxiliary test structures for process verification and possible spacing-dependent evaluation. The breakdown electric field (E_BR) was estimated as V_BR/L_gap and is referred to here as the average lateral breakdown field, because the actual electric field is nonuniform and can be enhanced near the electrode edges. Moreover, all measurements were conducted with the samples immersed in Fluorinert electronic liquid to prevent surface flashover in air at high voltages. And to ensure statistical validity and reproducibility, the lateral breakdown characteristics were measured across at least 10 different locations for each sample. The breakdown voltages reported in this work represent the average values.

3. Results

The breakdown voltage of all the aforementioned samples was tested using B1505A. The high-voltage (HV) module was configured to scan from 0 to 3000 V in forward bias and from 0 to −3000 V in reverse bias. The I-V scan was terminated when the current reached the device’s limit of 4 mA [24]. The I-V curve exhibited a distinct sharp increase in current as the applied voltage rose to a critical point. We defined the breakdown voltage of the sample as the point at which the leakage current density exceeded 1 × 10⁻⁶ A/mm².

3.1. The Effect of C Doping Concentration on the Breakdown Characteristics of cGaN Buffer

To systematically evaluate the impact of C doping concentration on the crystal quality of GaN buffer layers, we performed HRXRD measurements on all samples as shown in Figure 3. The FWHM values of (002) and (102) reflections serve as integral indicators of the crystalline quality, as they can be influenced by various factors including dislocations, point defects, and local strain variations. Threading dislocation densities were approximately estimated from the rocking-curve FWHM values measured by HRXRD. According to the GaN-specific methodology summarized by Moram and Vickers [25], the symmetric (002) reflection of c-plane GaN is mainly sensitive to screw-type threading dislocations, while the asymmetric (102) reflection is more suitable for evaluating edge-type threading dislocations. In this work, the dislocation density was calculated using the relation proposed by Kurtz et al. [26] for a random dislocation distribution as shown in Equation (1):

$$\rho ={\displaystyle \frac{{\beta }^2}{{9b}^2}}$$

(1)

Thus, the screw and edge type threading dislocation densities were estimated as

$${\rho }_{screw }=\frac{{\beta }_{\left(002\right)}^2}{9b_{screw}^2}$$

$${\rho }_{edge }={\displaystyle \frac{{\beta }_{\left(102\right)}^2}{9b_{edge}^2}}$$

$${\rho }_{total}={\rho }_{screw }+{\rho }_{edge }$$

The ρ_screw and ρ_edge are the density of screw and edge dislocations, respectively; β₍₀₀₂₎ and β₍₁₀₂₎ are the FWHM values of the rocking curves for the corresponding crystal planes (converted to radians); b_screw = <0001> (magnitude of 0.5185 nm) and b_edge = <1120> (magnitude of 0.3189 nm) are the Burgers vectors of screw and edge dislocations in wurtzite GaN, respectively. This method has been proven to be applicable to various GaN heteroepitaxial systems including GaN-on-sapphire, and is a reliable method for dislocation density characterization of GaN epitaxial films [27,28].

Since the rocking-curve broadening may also arise from other defects and strain fluctuations, the obtained value should be regarded as an approximate indicator of crystalline quality rather than a strictly exact dislocation density in our research.

Table 2. FWHM and dislocation density for samples A1–A5.

Sample	Nominal C Concentration (cm−3)	(002) (arcsec)	Screw Dislocations (cm−2)	(102) (arcsec)	Edge Dislocations (cm−2)	TDDs (cm−2)
A1	8 × 1017	321.6	1.01 × 108	500.2	6.46 × 108	7.47 × 108
A2	1.61 × 1018	362.6	1.28 × 108	524.6	7.11 × 108	8.39 × 108
A3	4.69 × 1018	394.3	1.51 × 108	566.7	8.29 × 108	9.81 × 108
A4	9.38 × 1018	468.2	2.13 × 108	629.9	1.02 × 109	1.24 × 109
A5	1.34 × 1019	560.8	3.06 × 108	743.2	1.43 × 109	1.73 × 109

presents the FWHM values of the XRD (002) and (102) reflections and the corresponding screw and edge dislocation densities derived from these measurements.

As shown in Table 2, the FWHM values of both crystallographic planes exhibit a monotonic increasing trend with increasing nominal C doping concentration. The FWHM of the (002) reflection increases progressively from 321.6 arcsec to 560.8 arcsec, representing a 74.3% increase; similarly, the (102) reflection FWHM increases from 500.2 to 743.2 arcsec, an increase of 48.6%. The estimated TDDs also rise from 7.47 × 10⁸ to 1.73 × 10⁹ cm⁻², with a total increase of 131.7%. These results clearly demonstrate that the structural metrics become worse monotonically when nominal carbon concentration increases, and excessive C doping concentration leads to severe deterioration of crystal quality.

AFM was used to assess the effect of C doping on the surface morphology of the cGaN buffer. As shown in Figure 4, the RMS roughness increases from 0.33 nm for A1 to 0.41 nm, 0.50 nm, 1.3 nm, and 2.1 nm for A2–A5, respectively. Only a slight increase in roughness is observed at low-to-moderate doping levels (A1–A3), indicating that the surface morphology is well maintained in this range. However, further increasing the C doping level to A4 and A5 results in a marked deterioration of the surface. This phenomenon, in which surface roughness increases monotonically with nominal doping concentration, is consistent with the results obtained from XRD testing. While the RMS roughness provides a general statistical measure of the surface, a closer inspection of the AFM topographical images (as shown in Figure 4) reveals a more complex hierarchical relief. As correctly pointed out, the surface morphology is characterized by the coexistence of large-scale wavy features and small island-like structures. The large wavy features are typically associated with the step-bunching phenomenon, which occurs due to the step-flow growth mode and localized strain relaxation during the epitaxial process. Superimposed on these steps are densely distributed nanometer-scale small islands. These islands likely originate from localized 3D nucleation or the termination points of threading dislocations at the surface.

HV measurements were carried out to evaluate the effect of C doping on the lateral breakdown behavior of the cGaN buffer layer. As shown in Figure 5, all samples exhibit very low leakage current (<10⁻⁸ A/mm²) at low bias, followed by a sharp current increase at breakdown. Among them, A3 shows the best performance, maintaining low leakage current up to 1920 V and exhibiting symmetric breakdown characteristics with absolute breakdown voltages (|V_BR|) both of 1920 V for reverse and forward biases, respectively. In contrast, A1 and A5 break down at significantly lower voltages, indicating that either insufficient or excessive C doping is unfavorable for breakdown performance. Figure 6b further confirms that A3 has the narrowest V_BR distribution, reflecting the best stability and reproducibility. The superior breakdown behavior of A3 arises from the synergistic optimization of compensation effect and crystal quality. In this work, V_BR was extracted using a current density of 1 × 10⁻⁶ A/mm², which was chosen to limit excessive self-heating during high-voltage measurements and to allow consistent comparison with previous reports. It should be noted that the physical onset of hard breakdown, where the I-V characteristic sharply transitions from trap-mediated leakage to catastrophic breakdown, occurs slightly earlier.

Notably, the (002) and (102) reflection FWHM values of sample A3 (nominal C doping concentration of 4.69 × 10¹⁸ cm⁻³) are 394.3 and 566.7 arcsec, respectively, with TDDs of 9.81 × 10⁸ cm⁻². It shows moderately higher FWHM and dislocation density than the lightly doped samples A1 and A2, but outperforms the heavily doped A4 and A5. At the same time, the RMS of A3 is 0.50 nm, which is slightly higher than that of A1 and A2 but still significantly lower than those of A4 and A5. Moreover, the AFM images reveal a relatively uniform surface with clear grain boundaries and limited defects in A3, in contrast to the rough island-like and grooved morphologies observed in A4 and A5. This is in good agreement with the crystal quality results and provides further support for the superior electrical performance of A3. This unique positioning enables sample A3 to simultaneously benefit from moderate C doping while successfully avoiding the severe crystal quality degradation associated with excessive doping, thereby establishing a critical physical foundation for its superior breakdown performance.

The increase in breakdown voltage resulting from moderate C doping is consistent with the enhanced buffer compensation induced by C incorporation, through which vertical and lateral leakage pathways are effectively suppressed. At the same time, its relatively low TDDs and smooth surface help minimize defect-related trap states and electric-field inhomogeneity, delaying the onset of localized current percolation. In contrast, underdoping fails to sufficiently suppress background conduction, while overdoping degrades the crystal quality and surface morphology, leading to enhanced trap-assisted leakage, local field concentration, and earlier breakdown. As a result, the optimum carbon concentration is process-dependent and should be considered together with temperature, pressure, V/III ratio, and layer thickness. The nominal C concentration of A3 is the most favorable in this series under the present growth conditions, which achieved a trade-off between insulation capability and structural quality, giving rise to the highest and most stable lateral breakdown voltage.

3.2. The Effect of Buffer Thickness on Breakdown Characteristics of cGaN Buffer

To further optimize the cGaN buffer, the influence of buffer thickness on crystal quality was examined while keeping the nominal C concentration constant at 4.69 × 10¹⁸ cm⁻³. HRXRD characterization was performed for samples B1–B4 with thicknesses ranging from 0.25 to 1.50 μm. As shown in Figure 7 and

Table 3. FWHM and dislocation density for samples B1–B4.

Sample	cGaN Thickness (μm)	(002) (arcsec)	Screw Dislocations (cm−2)	(102) (arcsec)	Edge Dislocations (cm−2)	TDDs (cm−2)
B1	0.25	617.9	3.72 × 108	643.2	1.07 × 109	1.44 × 109
B2	0.50	453.2	2.00 × 108	632.8	1.03 × 109	1.23 × 109
B3	1.00	449.6	1.97 × 108	582.3	8.76 × 108	1.07 × 109
B4	1.50	561.7	3.07 × 108	644.7	9.96 × 108	1.30 × 109

, the FWHMs of the (002) and (102) reflections first decrease and then increase with increasing thickness, revealing a non-monotonic thickness dependence of crystalline quality. Among all samples, B3 exhibits the narrowest diffraction linewidths, with minimum FWHMs of 449.6 arcsec for (002) and 582.3 arcsec for (102).

This evolution is consistent with the dislocation density analysis. For a relatively thin cGaN buffer such as B1 (TDDs ≈ 1.44 × 10⁹ cm⁻²), the accommodation of lattice and thermal mismatch between GaN and sapphire may be insufficient, which can result in incomplete strain relaxation, less effective dislocation reduction, and degraded crystalline quality. Increasing the cGaN thickness can improve strain accommodation and defect reduction up to an optimized thickness under the growth conditions, thereby improving the crystal quality, which can be inferred from the reduction in FWHM and dislocation density. For example, B3 achieves the lowest TDDs of 1.07 × 10⁹ cm⁻².

In contrast, when the buffer thickness is further increased to 1.50 μm, the crystal quality deteriorates again. This may be related to changes in growth kinetics caused by an excessively thick buffer layer. Under optimized epitaxial conditions, GaN is expected to grow in an ideal 2D step-flow mode. However, during the prolonged epitaxy required for very thick layers, maintaining the strictly optimal growth window becomes challenging. The reduction in Ga adatom mobility disrupts the steady step-flow growth, promoting step bunching and macro-step formation, which manifests as the increased surface roughness observed in our samples. Furthermore, severe deviations can trigger a transition from 2D growth to 3D growth features, such as hillocks or microscopic faceting. Macroscopically, these morphological degradations induce severe structural instabilities, leading to the formation of new grain or tilt boundaries and the generation of secondary threading dislocations. This can also be verified by the subsequent AFM results.

AFM characterization further corroborates the non-monotonic dependence of crystal quality on buffer thickness revealed by XRD. As shown in Figure 8, the thinnest sample B1 shows pronounced surface undulations and island-like features (RMS ≈ 0.95 nm), consistent with its high dislocation density and broad diffraction peaks. As the buffer thickens to intermediate values (0.50 μm/1.00 μm), opportunities for dislocation interaction, bending, and annihilation increase, leading to markedly improved surface planarity. Sample B3 attains the lowest RMS (≈0.37 nm) and the minimum TDDs, showing that this thickness effectively balances strain relaxation with suppression of defect propagation. This indicates that an insufficient buffer thickness cannot adequately relieve lattice and thermal mismatch, allowing threading dislocations to reach the surface, impede adatom mobility, and disrupt step-flow growth, which together cause surface roughening.

Further increasing the buffer to 1.50 μm degrades both surface and crystal quality. This reversal can be attributed to accumulated strain energy and a prolonged growth/thermal history in thicker layers, which promote new defect formation, local stress concentration, and a transition toward more three-dimensional growth modes that suppress favorable step-flow kinetics. Therefore, the observed correspondence between AFM and XRD demonstrates that the surface instabilities may further generate structural imperfections, including grain or tilt boundaries and additional threading dislocations, which in turn deteriorate the electrical performance by increasing leakage pathways. Thus, the degradation at larger cGaN thicknesses is more reasonably attributed to non-optimal growth kinetics and morphology-induced defect formation, rather than to continued strain relaxation alone. It is worth noting that the non-monotonic dependence of dislocation density on buffer thickness may also be affected by the growth rate, which warrants further investigation in future work.

The lateral breakdown behavior was evaluated by HV measurements in a horizontal device geometry (Figure 9a,b). Breakdown strongly depends on the cGaN buffer thickness. Under reverse bias, the thinnest sample in this series B1 (0.25 μm) with the highest TDDs (≈1.44 × 10⁹ cm⁻²) and roughest surface (RMS ≈ 0.95 nm) shows the poorest blocking (|V_BR| = 1830 V). Increasing the thickness to B2 (0.50 μm) yields a modest improvement (|V_BR| = 1950 V) associated with reduced TDDs and smoother surface. The best reverse performance occurs at B3 (1.00 μm), which reaches 2340 V and coincides with the lowest TDDs (≈1.07 × 10⁹ cm⁻²) and minimal RMS (0.37 nm). Further thickening to B4 (1.50 μm) degrades the blocking ability (|V_BR| = 2040 V) as TDDs and surface roughness rise again. The forward bias trends mirror the reverse case: B3 yields the highest forward breakdown (|V_BR| = 2310 V), while both thin and over-thick buffers show reduced forward blocking (|V_BR| = 1770 V for B1; |V_BR| = 1880 V for B4). This bilateral behavior implies that buffer quality—not bias polarity—controls lateral breakdown.

Mechanistically, the thickness dependence reflects a trade-off between strain relaxation and defect generation. An insufficient buffer thickness allows threading dislocations to propagate to the surface, creating leakage paths and local field-enhancement sites that trigger premature impact ionization and breakdown. Moderate thickness (around 1.00 μm) promotes dislocation bending, interaction and partial annihilation during growth, yielding fewer conductive defect channels, more uniform field distribution, and improved surface planarity—together delaying avalanche onset. Excessive thickness, however, accumulates strain energy and extends the growth/thermal history, which can nucleate new defects, enhance local stress concentrations, and induce more three-dimensional growth modes; these effects revive defect-assisted leakage and lower the effective blocking voltage.

In summary, the lateral breakdown voltage exhibits a pronounced non-monotonic dependence on the cGaN buffer thickness. Under our present growth window, the highest and most symmetric breakdown performance is achieved at approximately 1.00 μm, a point where defect annihilation is maximized before the onset of growth kinetic degradation. Rather than indicating an absolute thickness limit, the deterioration in thicker layers emphasizes the necessity of maintaining stable step-flow growth kinetics during prolonged epitaxy. More importantly, realizing high-voltage cGaN buffer layers requires coupling these stabilized growth conditions with a meticulously optimized nominal carbon concentration window. While sufficient carbon incorporation is crucial to compensate background shallow donors and increase the buffer resistivity, excessive carbon doping must be avoided to prevent introducing additional defect-related penalties that would otherwise compromise the device performance.

3.3. The Effect of AlN Nucleation Layer Thickness on Breakdown Characteristics of the cGaN Buffer

The AlN nucleation layer thickness strongly controls cGaN crystal quality via dislocation filtering and strain management mechanisms. Four samples with AlN thicknesses of 320 nm, 400 nm, 480 nm and 640 nm were compared by HRXRD (Figure 10) and the best crystal quality was obtained near 480 nm as shown in

Table 4. FWHM and dislocation density for samples C1–C3 and B3.

Sample	AlN Thickness (nm)	(002) (arcsec)	Screw Dislocations (cm−2)	(102) (arcsec)	Edge Dislocations (cm−2)	TDDs (cm−2)
C1	320	550.6	2.95 × 108	598.8	9.26 × 108	1.22 × 109
B3	400	449.6	1.97 × 108	582.3	8.76 × 108	1.07 × 109
C2	480	439.6	1.88 × 108	570.1	8.39 × 108	1.03 × 109
C3	640	470.4	2.15 × 108	590.8	9.01 × 108	1.12 × 109

. A too thin AlN layer seemingly cannot sufficiently accommodate lattice mismatch or induce the bending/trapping of threading dislocations at the substrate–buffer interface, so defects readily propagate into the GaN (highest TDDs observed for 320 nm). Moderately thick AlN (around 400 nm) already enhances elastic strain relief and promotes dislocation interaction and annihilation, improving FWHM and reducing TDDs. Extending thickness to 480 nm further approaches the ideal balance: the nucleation layer provides enough thickness to redistribute mismatch stress and generate favorable glide/bending of dislocations without introducing new intrinsic defects or interface roughness, yielding the minimum measured TDDs. Beyond this practical thickness range (e.g., above 640 nm in our samples), increased AlN thickness tends to correlate with increased defect generation and surface roughening. This may be attributed to a combination of factors, potentially including accumulated tensile stress within the AlN layer and a transition towards less ideal, more three-dimensional overgrowth modes, which can reintroduce stress concentrations and facilitate new defect nucleation, thus diminishing the structural and electrical benefits.

In short, within the investigated range, the AlN nucleation layer shows a non-monotonic influence on crystal quality. The intermediate thickness of 480 nm gives the lowest XRD rocking-curve FWHM and the smoothest AFM surface among the samples studied, indicating improved crystalline quality and reduced surface roughness. Both thinner and thicker AlN layers lead to inferior results. Combining the optimized AlN layer with the previously identified cGaN buffer (1.00 μm) and nominal C doping concentration yields the best overall structural performance in this sample set.

AFM results, as shown in Figure 11, indicate that the cGaN layer surface morphology depends on the AlN nucleation layer thickness within the present growth window. For the thinner AlN layer (320 nm), the surface morphology is comparatively rough, which is consistent with insufficient accommodation of lattice and thermal mismatch near the AlN/GaN interface. In this case, threading dislocations may propagate more readily into the GaN layer, leading to dislocation-related growth perturbations such as hillock formation and irregular terraces, as reflected by the higher roughness (RMS ≈ 0.55 nm). By contrast, when the AlN thickness increases to an intermediate range (e.g., 400~480 nm), the surface becomes noticeably smoother. The 480 nm sample, in particular, shows the clearest atomic terraces and the lowest roughness (RMS ≈ 0.30 nm). This trend suggests that the intermediate AlN thickness provides a more favorable growth template, which may help suppress dislocation-related surface disturbances and promote a more uniform step-flow-like growth mode.

However, further increasing the AlN thickness to 640 nm degrades the morphology again (RMS ≈ 0.82 nm), indicating that the benefit of a thicker nucleation layer is not unlimited. Once the AlN layer becomes excessively thick, accumulated internal stress within the AlN layer itself may no longer be efficiently accommodated and may instead influence the GaN overlayer in a less uniform manner. At the same time, an overly thick AlN template may be more susceptible to additional structural imperfections or interfacial roughness, which would weaken its role as an ideal epitaxial template. These effects may destabilize step-flow growth and promote local three-dimensional nucleation or mound formation during the early GaN growth stage, leading to renewed roughening.

Overall, an intermediate thickness near 480 nm provides the most favorable balance within the present growth window: it is thick enough to improve the surface roughness and the step-flow growth pattern of the material and suppress the vertical propagation of threading dislocations, but not so thick that the AlN layer itself appears to introduce additional stress concentration or structural degradation. This interpretation is consistent with the observation that the smoothest surface is obtained only within a limited thickness window rather than by continuously increasing the AlN thickness.

Lateral electrical testing shows a clear correlation between AlN nucleation thickness, buffer quality, and high-voltage performance in Figure 12. The best lateral breakdown is observed for sample C2 (AlN = 480 nm), with symmetric forward and reverse |V_BR| of 2850 V and the lowest mid-to-high voltage leakage (10⁻⁹~10⁻⁸ A/mm²). Considering the 30 μm gap distance, this translates to a lateral E_BR of approximately 0.95 MV/cm. This behavior is consistent with the improved structural quality of this sample, including the reduced threading dislocation density and the smoother surface morphology. A lower density of extended defects and a flatter surface are both expected to reduce local current leakage pathways and mitigate field enhancement at morphological irregularities. In addition, a more structurally uniform buffer may support a more homogeneous internal field distribution, which can help delay electrical failure under high bias. Together, these factors are consistent with the enhanced breakdown performance and reduced leakage observed for the intermediate-thickness sample.

By contrast, both the thin sample C1 (AlN = 320 nm) and the overly thick sample C3 (AlN = 640 nm) show reduced |V_BR| and higher leakage. For the thin case, insufficient strain accommodation may leave more threading dislocations unfiltered, and the associated defects may contribute to local leakage conduction and earlier electrical failure. For the excessively thick case, accumulated internal stress, degraded AlN layer quality, and possible interfacial roughness may introduce additional localized high-field regions and defect-related leakage paths, which can weaken the lateral blocking capability. The intermediate sample B3 (AlN = 400 nm) shows intermediate electrical performance, consistent with a gradual improvement in buffer quality as thickness approaches the favorable range.

The close forward/reverse symmetry of |V_BR| across samples indicates that breakdown is likely governed by the overall buffer quality and field distribution rather than by a strong polarity-dependent interface effect. Small forward/reverse differences in individual devices likely reflect local microstructural variation or electrode-edge field effects rather than a systematic polarity dependence. In summary, AlN thickness tuning within the present growth window has a strong influence on electrically active defects, surface morphology, and field distribution, thereby affecting lateral blocking capability; deviations on either side of the favorable thickness range are associated with increased leakage and lower |V_BR|.

3.4. Preliminary Investigation of Low Pressure and Reduced V/III Treatments on Breakdown Characteristics of the Buffer Performance

Building on the optimized baseline C2, we further investigated the influence of growth pressure and V/III ratio on the crystalline quality and breakdown characteristics of cGaN. By reducing the growth pressure from 300 to 50 mbar (sample D1) and subsequently lowering the NH₃ flow rate (sample D2), a progressive narrowing of HRXRD (002) and (102) rocking curves was observed as shown in Figure 13, corresponding to a monotonic decrease in TDDs from 1.03 × 10⁹ cm⁻² in C2 to 8.38 × 10⁸ cm⁻² in D2 (

Table 5. FWHM and dislocation density for samples C2 and D1–D2.

Sample	cGaN Pressure (mbar)	cGaN V/III Ratio	(002) (arcsec)	Screw Dislocations (cm−2)	(102) (arcsec)	Edge Dislocations (cm−2)	TDDs (cm−2)
C2	300	1739	439.6	1.88 × 108	570.1	8.39 × 108	1.03 × 109
D1	50	1739	413.5	1.66 × 108	545.6	7.69 × 108	9.35 × 108
D2	50	870	388.4	1.47 × 108	517.4	6.91 × 108	8.38 × 108

).

The observed improvements arise from the synergistic modification of surface kinetics and precursor chemistry. Lowering the reactor pressure increases the mean free path of Ga adatoms, which effectively extends their diffusion length. This promotes a 2D step-flow growth mode, suppressing 3D island nucleation—a primary source of threading dislocations—and facilitating the lateral annihilation of existing dislocations. Concurrently, a reduced V/III ratio minimizes kinetic constraints by decreasing surface competition from excess nitrogen species and suppressing parasitic gas-phase prereactions. The resulting increase in Ga adatom residence time further enhances lateral incorporation at step edges. These mechanisms—pressure-driven diffusion enhancement and V/III ratio-driven barrier reduction—collectively bias the growth process toward defect-reducing surface kinetics. These results demonstrate that when implemented on a well-engineered heterointerface, sequential optimization of these growth parameters provides a scalable pathway for defect suppression in cGaN epitaxy.

AFM characterization reveals progressive surface refinement across the optimization series as shown in Figure 14. Sample C2 exhibits RMS roughness of 0.30 nm with characteristic undulating morphology. Sample D1, grown at reduced pressure (50 mbar), shows improved surface flatness with RMS roughness of 0.26 nm, reflecting enhanced step-flow growth kinetics from increased Ga adatom surface diffusion. Lower pressure reduces interstep distances and promotes uniform lateral growth, minimizing 3D island formation.

Sample D2, with additionally reduced NH₃ flow rate, achieves the lowest RMS roughness of 0.20 nm, demonstrating the cumulative benefit of dual optimization. The superior surface flatness originates from synergistic interplay between low-pressure enhancement of Ga adatom migration and reduced V/III ratio alleviation of kinetic constraints on lateral diffusion, both promoting optimal 2D step-flow growth. Mechanistically, diminished NH₃ flow suppresses parasitic side reactions and reduces surface contamination, enabling cleaner, more ordered growth surfaces. Additionally, lower V/III ratio reduces nitrogen-induced adsorption competition, prolonging Ga residence time and facilitating step-edge incorporation that favors smooth terrace formation over island nucleation.

The surface morphology progression directly correlates with XRD-derived defect densities: smoother surfaces reflect superior crystalline ordering, as reduced roughness minimizes nucleation sites for defect formation during growth. The inverse relationship between surface roughness and dislocation density confirms that step-flow growth geometry—enabled by optimized surface kinetics—suppresses dislocation generation and propagation. Collectively, decreasing FWHM, TDDs and surface roughness demonstrate that sequential parameter optimization synergistically enhances cGaN buffer structural quality through complementary mechanisms governing surface diffusion kinetics and growth mode selection. But practical limits remain: overly low pressure or V/III can upset stoichiometry or introduce other defects, so the observed improvements pertain to the explored parameter window.

Electrical characterization of samples reveals progressive performance enhancement correlating with structural quality improvements as shown in Figure 15. Sample C2 exhibits |V_BR| of 2850 V for reverse bias with elevated reverse leakage current (~1 × 10⁻⁸ A/mm² at high bias), attributable to its higher TDDs (1.03 × 10⁹ cm⁻²) and rougher surface (RMS ≈ 0.30 nm).

Sample D1, incorporating reduced-pressure growth, shows marginal improvement: |V_BR| of 2940 V (+3.2% vs. C2) with reduced reverse leakage current correlated with lower TDDs (9.35 × 10⁸ cm⁻²) and improved surface (RMS ≈ 0.26 nm). This modest enhancement reflects the limited effectiveness of structural optimization alone in suppressing defect-mediated conduction. Sample D2, with combined reduced pressure and reduced NH₃ flow, exhibits exceptional performance: no measurable breakdown within the testing limit and leakage current suppressed 1~2 orders of magnitude below C2 and D1 across the entire voltage range. The measured breakdown voltage of >3000 V corresponds to a breakdown electric field of approximately >1.0 MV/cm for the given electrode spacing of 30 μm.

Forward bias characteristics (Figure 15b) corroborate these findings: C2 exhibits 2850 V forward breakdown with steeply rising forward leakage, D1 shows 2970 V (+4.2%) with improved suppression and D2 shows no breakdown within the measurement window with dramatically suppressed forward leakage, reflecting symmetric behavior consistent with bulk-limited conduction in a semi-insulating material.

The progression from C2 to D1 demonstrates that pressure reduction alone yields consistent but modest improvement through enhanced surface diffusion and reduced dislocations (Figure 16a). However, the extraordinary leap from D1 to D2 as shown in Figure 16b—manifested in breakdown suppression and 1~2-order-of-magnitude leakage reduction—may be related to the combined effect of improved structural quality and modified compensation conditions under the reduced V/III ratio. Based on our previous SIMS calibration for the GaN growth window, a reduced V/III ratio may favor stronger carbon incorporation under similar growth conditions, which could in turn modify the compensation state and influence the leakage behavior. At the same time, practical caveats are also important that excessive carbon or too low V/III can introduce deep compensating states distributed unfavorably or produce stoichiometric imbalance and new defect species; similarly, extreme pressure reduction may destabilize growth uniformity. Thus the pronounced enhancement observed for D2 pertains to the optimized window explored here, where kinetic and compositional tuning are relatively balanced to minimize active defects and maximize bulk resistivity.

4. Conclusions

In this work, we successfully implemented a multi-parameter co-optimization strategy for C-doped GaN buffer layers grown on sapphire substrates via MOCVD, aiming to achieve high lateral breakdown voltage for power electronic applications. Through a systematic investigation of C doping concentration, cGaN buffer layer thickness, AlN nucleation layer thickness and cGaN growth pressure and V/III ratio, we established critical correlations between epitaxial growth parameters, material quality, and the resulting electrical performance.

The present study demonstrates a GaN buffer layer on sapphire with a high buffer-level breakdown voltage exceeding 3000 V in the investigated lateral test structure, without the aid of device-engineering features such as field plates, indicating improved high-voltage blocking capability of the epitaxial structure. This result is accompanied by a low threading dislocation density and excellent surface roughness, reflecting the effectiveness of the growth-optimization strategy used in this work and providing a useful material basis for further development of GaN-on-sapphire power electronics.